激光与光电子学进展, 2019, 56 (1): 010702, 网络出版: 2019-08-01
基于核相关滤波的长期目标跟踪算法 下载: 1157次
Long-Term Object Tracking Algorithm Based on Kernelized Correlation Filter
傅里叶光学 目标跟踪 核相关滤波 快速角点检测 双向光流 遮挡 Fourier optics target tracking kernelized correlation filter fast corner detection bidirectional optical flow occlusion
摘要
针对传统核相关滤波器(KCF)无法处理严重遮挡及光照变化等问题,提出一种结合快速角点检测与双向光流法的长期KCF跟踪算法。首先利用KCF跟踪器在目标位置上提取融合方向梯度直方图特征、颜色属性特征和灰度特征的多通道特征,计算输出响应图并得到所跟踪目标的峰值旁瓣比(PSR),然后通过比较PSR与经验阈值来判断目标是否被遮挡;当目标出现遮挡时,在快速角点检测的角点基础上利用双向光流法重新检测下一帧目标位置,并采用一种新模板更新策略来应对严重遮挡。与其他算法进行对比实验,验证了本文算法对处理遮挡和光照变化具有高效性及稳健性。
Abstract
Focusing on the issue that the traditional kernelized correlation filter (KCF) has poor performance in handing heavy occlusion and illumination variations, a long-term KCF tracking algorithm is proposed combined with fast corner detection and bidirectional optical flow method. First, the KCF tracker is used to extract the multi-channel features of the histogram of gradient, color attributes, and gray features at the target location. The output response map is calculated and the peak sidelobe ratio (PSR) of the tracked target is obtained. The PSR and the empirical threshold determine whether the target is occluded by comparison. When the target is occluded, the bidirectional optical flow method is used to redetect the target position of the next frame based on the corner points detected by the fast corner detection, and a new template updating strategy is adopted to deal with the heavy occlusion. Compared with other algorithms, the proposed algorithm is effective and robust to the processing of occlusion and illumination variations.
茅正冲, 陈海东. 基于核相关滤波的长期目标跟踪算法[J]. 激光与光电子学进展, 2019, 56(1): 010702. Zhengchong Mao, Haidong Chen. Long-Term Object Tracking Algorithm Based on Kernelized Correlation Filter[J]. Laser & Optoelectronics Progress, 2019, 56(1): 010702.